[ROCm] preshuffled weight mm #1702
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/1702
Note: Links to docs will display an error until the docs builds have been completed. ❌ 29 New FailuresAs of commit 8b57424 with merge base 70fc520 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
albanD
reviewed
Feb 13, 2025
| from torchao.swizzle.swizzle_tensor import SwizzleTensor | ||
|
|
||
| aten = torch.ops.aten | ||
| SWIZZLE_OPS_TABLE: Dict[Any, Any] = {} |
There was a problem hiding this comment.
Given how you use this, won't you be able to use the builtin version of this: https://pytorch.org/docs/stable/library.html#torch.library.register_torch_dispatch
|
@jcaip has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@mxz297 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
1 similar comment
|
@mxz297 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@mxz297 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
Note that other extensions aren't compiling currently but weren't added by this PR. I've had to modify conflicts in setup.py now that certain extensions are being built for rocm target. swizzle extension if compiled by itself is building and working fine. |
|
Had to apply this diff to this PR to get setup.py to work on rocm. Comment out sparse_marlin and other extension source files and only leave swizzle alone. diff --git a/setup.py b/setup.py
index c181267f..3d10de5c 100644
--- a/setup.py
+++ b/setup.py
@@ -350,15 +350,15 @@ def get_extensions():
extensions_hip_dir = os.path.join(
extensions_dir, "cuda", "tensor_core_tiled_layout"
)
- hip_sources = list(
- glob.glob(os.path.join(extensions_hip_dir, "*.cu"), recursive=True)
- )
- extensions_hip_dir = os.path.join(extensions_dir, "cuda", "sparse_marlin")
- hip_sources += list(
- glob.glob(os.path.join(extensions_hip_dir, "*.cu"), recursive=True)
- )
+ #hip_sources = list(
+ # glob.glob(os.path.join(extensions_hip_dir, "*.cu"), recursive=True)
+ #)
+ #extensions_hip_dir = os.path.join(extensions_dir, "cuda", "sparse_marlin")
+ #hip_sources += list(
+ # glob.glob(os.path.join(extensions_hip_dir, "*.cu"), recursive=True)
+ #)
extensions_hip_dir = os.path.join(extensions_dir, "rocm")
- hip_sources += list(
+ hip_sources = list(
glob.glob(os.path.join(extensions_hip_dir, "**/*.hip"), recursive=True)
)
hip_sources += list( |
|
@mxz297 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
mxz297
approved these changes
Apr 8, 2025
|
seems the PR breaks ruff lint checks and test-mps-ops (macos-m1-stable) |
jerryzh168
added a commit
that referenced
this pull request
Apr 9, 2025
This reverts commit 5abaa35.
|
reverting, please fix ruff before landing: https://hud.pytorch.org/pr/pytorch/ao/1702#40204953928 |
jerryzh168
added a commit
that referenced
this pull request
Apr 9, 2025
|
@jeffdaily it seems like in test-mps-ops (macos-m1-stable), we still have swizzle source file used for compilation? The ruff code analysis failures also needs to be addressed. |
|
New PR here #2044 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Adds SwizzleTensor subclass that wraps a Tensor and reorders the contents to be suitable for HIPBLASLT_ORDER_COL16_4R8. SwizzleTensor intercepts torch.mm and replaces with custom calls to hipblaslt.